AI Engineering · ENGLISH · ONE-DAY WORKSHOP

Build an Agent-Powered App with Claude

Agent Harness Engineering: Build the application around the agent.

An agent application needs to manage more than model responses. It must handle tool execution, show progress, request approval, preserve session state and make completed work accessible.

Explore the sample outcome View the day’s agenda

29 November 2026

Melbourne · Proposed: 10:00 AM–4:00 PM AEDT
Includes lunch break and two short breaks. Meals are not advertised as included.

Early bird A$328 · Standard A$389

Standard fee: A$389. Early bird: A$328 until 22 November 2026, 11:59 PM AEDT (Melbourne time).

Instructor and exact venue: to be confirmed. Account requirements and any software or API charges will be confirmed before registration.

See the kind of outcome you’ll build

Sample deliverable: client website review

Fictional, pre-generated output. This preview does not run an agent. The full workshop reference application is still being prepared.

Brief: Studio North needs a service page that explains a website review to independent consultants.

Findings

The supplied fictional brief identifies the audience and review outputs but gives no delivery timeframe or confirmed price. The page should state the review scope and invite a scope discussion; it must not invent those missing details.

Recommended page

Headline: Make your consulting service easier to understand. Deliverables: an annotated page review, suggested structure and prioritised revisions. Exclusions: implementation, hosting and guaranteed enquiries. Next step: share a public URL and the service you want to explain.

Review status

Draft complete. Human approval required before treating this as a final client deliverable. Missing facts remain visibly unresolved.

About the workshop

Build a focused web application around a Claude agent runtime. The teaching project is a Client Deliverable Workspace: a user provides a task and sample project documents; the agent reads the materials, prepares a deliverable and requests approval before saving its final version.

The user can inspect tool activity, open the completed document and return to a saved session. Work from a prepared application scaffold and implement the key connections between the interface, runtime, tools and session store.

What you’ll learn

1. Understand the responsibilities of the interface, backend, agent runtime and execution environment.

2. Translate runtime events into streaming responses, tool activity and execution states.

3. Give the agent controlled access to a project workspace.

4. Implement approval, rejection and cancellation behaviour.

5. Save conversation and execution records and reopen a session.

6. Make complete generated documents accessible through the application.

7. Test tool errors, interruption and incorrect completion claims.

Your day, session by session

Five hours of instruction and practical work within a six-hour day. Times are proposed.

10:00–10:30 · Explore the Completed Agent App

Run a task, inspect tool activity, approve an action and open the finished deliverable.

10:30–11:30 · Connect the App to the Runtime

Trace the scaffold architecture and implement the request and event flow.

11:30–11:45 · Break

11:45–12:45 · Build Streaming and Tool Visibility

Render response events, tool calls, results and execution states.

12:45–13:15 · Lunch

13:15–14:30 · Add Workspace Tools and Approval

Connect restricted file operations, implement an approval checkpoint and preview a generated document.

14:30–14:45 · Break

14:45–15:40 · Save Sessions and Test Failures

Reopen a session and test rejected actions, tool failures and interrupted execution.

15:40–16:00 · Application Demo and Architecture Review

Demonstrate the full user journey and explain remaining engineering limitations.

What you’ll leave with

A locally runnable agent-powered web application.
A streaming interface with visible tool activity.
A restricted project workspace and human approval checkpoint.
Saved session records and a basic reopen flow.
A complete sample document accessible inside the application.
A failure-testing checklist and architecture notes.

A prepared scaffold supports a single-user local environment. Production authentication, multi-tenancy, billing, deployment and robust recovery of arbitrary in-flight tasks are outside scope.

Who should attend?

Developers and AI engineers comfortable with JavaScript or TypeScript, HTTP APIs and asynchronous programming. Frontend experience is helpful. Basic agent and tool-use knowledge is expected; attendance on 15 November is optional.

Before you attend

Bring a laptop with a code editor, Git and the specified Node.js runtime. Download the starter repository and complete the setup check in advance. The SDK version, runtime access requirements and responsibility for usage charges will be confirmed before registration.

Each workshop is standalone. No software credits, official certification or exam fees are advertised as included.

Other November workshops

8 November 2026 · Business AI